Updated: 2026-05

1. About This Page

Why is Comfy Cloud / ComfyUI node-based? I will outline its educational benefits by comparing it with other UIs that support Stable Diffusion (such as Web UI, Forge, Fooocus, and DALL-E built into ChatGPT).

If you have a teacher who uses TouchDesigner in class, you’ll probably think, “Oh, so it’s based on the same concept,” and it’ll all make sense.

2. The Three Layers of the UI

The UI of image-generating AI has about three levels of abstraction.

Level	Example	Features
High (Consumer)	ChatGPT, DALL-E, Midjourney, Adobe Firefly	Just write a prompt. The inner workings are a complete black box
Medium (Prosumer)	Stable Diffusion Web UI, Forge, Fooocus	Adjust parameters using sliders. Some customization possible
Low (Creator/Developer)	ComfyUI / Comfy Cloud	Internal processing is visible as nodes. Fully customizable

Comfy Cloud, which we cover in class, is the lowest layer. While it requires the most effort to learn, it offers the greatest insight into its inner workings.

3. Advantages of a Node-Based Approach

3.1 Visibility

Nodes and lines represent what is happening at each stage of the diffusion process. You can visually confirm that “the image is being reconstructed from the latent space via VAE decoding.” In a web UI, this occurs within a black box, so it is not visible to beginners.

That is the extent of its value as a teaching resource.

3.2 Flexible Reconfiguration

• “Inserting ControlNet” = Inserting a ControlNet node
• “Applying LoRA” = Adding a LoRA node
• “Refining intermediate latent images with a different K-Sampler” = Arranging K-Samplers in series

Everything is achieved by adding, removing, and reconfiguring nodes. Because the underlying mechanism is clear, it’s highly adaptable.

3.3 The Same Approach Can Be Applied to Video, 3D, and Audio

ComfyUI’s node-based architecture works with the same user experience not only for image generation but also for video (Wan, AnimateDiff), 3D (Hunyuan 3D), and audio generation (ACE Step). This allows training costs to be leveraged across different domains.

4. Disadvantages of a Node-Based Approach

I’ll be honest.

• It can be intimidating for beginners. You need to understand the nodes before you can even get to the point of “entering a prompt and generating output.”
• Mistakes in workflow setup can cause it to stop working (e.g., missing output nodes, incorrect wiring).
• There is too much freedom in choosing models and samplers, which often leads to confusion.
• While there is a culture of sharing finished workflows, it takes time to decipher other people’s workflows.

In class, we’ll start with a template provided by the teacher and gradually move toward creating our own designs.

5. Similarities and Differences with TouchDesigner

If you’ve already taken Professor Naka-Yasu’s TouchDesigner course, you’ll notice that there are many similarities.

• Place nodes on the screen and connect them with lines
• Distinguish pin colors by data type
• View results in real time

The difference is:

• TouchDesigner is designed to run every frame (real-time video)
• ComfyUI processes the data once when the run button is pressed (batch processing)

Even so, you can still use that same sense of visually tracking the flow of data.

6. Basics of Reading Nodes

When reading ComfyUI nodes, keep the following in mind.

1. What comes into the left input pin (follow the line)
2. What the node itself does (infer from the title + practice)
3. What comes out of the right-side output pin (what is passed to the next node)
4. Use line colors to identify the type of data flowing through (Model = purple, CLIP = yellow, latent image = pink, pixel image = blue)

At first, you check each one individually, but once you get the hang of it, you’ll be able to read the flow with your peripheral vision.

7. The “Correctness” and “Beauty” of Workflows

There is no “official correct answer” in ComfyUI. Even for workflows that produce the same results, the placement and order of nodes vary from person to person.

In class, just think of it as “if it works, it’s right” or “if it’s easy to read, even better.”

• Align elements in a top-to-bottom or left-to-right flow
• Place nodes with similar functions close together
• Use memo nodes to label sections

It’s just a matter of getting used to it. It’s okay if it’s a bit rough at first.

8. What’s Next

• Getting Started — How to Get Started with Comfy Cloud
• Minimum Workflow — Breaking Down the Minimum Workflow
• Parameters — Experimenting with Parameters